Deep-Learning-based Electrical Fault Detection in Photovoltaic Modules through Aerial Infrared Imaging: Addressing Data Complexity

Abstract

Aerial infrared imaging has emerged as areliable, efficient and promising technology for detecting electrical faults in photovoltaic modules. This is attributed to its non-invasive nature and capability to capture thermal signatures associated with defective components in large solar farms, that can be inspected ina fraction of the time required for ground-based methods. Nevertheless, the effectiveness of aerial infrared imaging in fault detection encounters complexities in the problem data representativeness, attributed to diverse conditions, such as module types and configurations, fault types, and even the acquisition environment, such as ambient temperature and humidity, irradiance levels, and wind conditions. This work presents the use of deep learning for electrical fault detection in photovoltaic modules while analyzing the inherent data complexity. This study explore the role of data complexity in influencing the performance of fault detection algorithms, high lighting the need for representative, consistent and balanced datasets encompassing diverse and real word fault scenarios.